# RoboOI Platform

> app.robooi.bio is where robotics teams rent time on a biocomputer (living neurons grown on a multi-electrode array) to train a robot skill in a closed loop, then download a policy that was validated against negative controls. Tagline: "Train with biology. Deploy on silicon." Company site: https://robooi.bio/

## How a training run works
1. Pick a robot: humanoid (Unitree G1), arm (Franka Panda), quadruped (Unitree Go2), dexterous hand (Shadow Hand) or drone (Skydio X2).
2. Pick a skill template, or upload demonstrations (CSV or JSON joint trajectories). Held poses in the demonstration become target poses, one neural pathway each.
3. Choose a biocomputer: a free simulated sandbox culture, a CL1 device, or a real recording (watch-only).
4. Calibrate: every electrode is probed with single pulses to find sites the culture answers on.
5. Train: closed-loop episodes. Success is followed by predictable feedback, misses by unpredictable feedback at random sites (protocol after Kagan et al., Neuron 2022).
6. Validate: the skill counts only if it passes with stimulation and fails under two negative controls (no stimulation, spikes blocked).
7. Download the policy: skill spec, decoder wiring, learning record and validation result.

## Honesty notes
- In this preview build the cultures are simulated; the protocol is the one used on real devices.
- The neurons are a training instrument and stay in the lab; robots run ordinary software.